You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

sla_porfsx_extended.c 39 kB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219
  1. #include <math.h>
  2. #include <stdlib.h>
  3. #include <string.h>
  4. #include <stdio.h>
  5. #include <complex.h>
  6. #ifdef complex
  7. #undef complex
  8. #endif
  9. #ifdef I
  10. #undef I
  11. #endif
  12. #if defined(_WIN64)
  13. typedef long long BLASLONG;
  14. typedef unsigned long long BLASULONG;
  15. #else
  16. typedef long BLASLONG;
  17. typedef unsigned long BLASULONG;
  18. #endif
  19. #ifdef LAPACK_ILP64
  20. typedef BLASLONG blasint;
  21. #if defined(_WIN64)
  22. #define blasabs(x) llabs(x)
  23. #else
  24. #define blasabs(x) labs(x)
  25. #endif
  26. #else
  27. typedef int blasint;
  28. #define blasabs(x) abs(x)
  29. #endif
  30. typedef blasint integer;
  31. typedef unsigned int uinteger;
  32. typedef char *address;
  33. typedef short int shortint;
  34. typedef float real;
  35. typedef double doublereal;
  36. typedef struct { real r, i; } complex;
  37. typedef struct { doublereal r, i; } doublecomplex;
  38. #ifdef _MSC_VER
  39. static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
  40. static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
  41. static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
  42. static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
  43. #else
  44. static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
  45. static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
  46. static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
  47. static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
  48. #endif
  49. #define pCf(z) (*_pCf(z))
  50. #define pCd(z) (*_pCd(z))
  51. typedef blasint logical;
  52. typedef char logical1;
  53. typedef char integer1;
  54. #define TRUE_ (1)
  55. #define FALSE_ (0)
  56. /* Extern is for use with -E */
  57. #ifndef Extern
  58. #define Extern extern
  59. #endif
  60. /* I/O stuff */
  61. typedef int flag;
  62. typedef int ftnlen;
  63. typedef int ftnint;
  64. /*external read, write*/
  65. typedef struct
  66. { flag cierr;
  67. ftnint ciunit;
  68. flag ciend;
  69. char *cifmt;
  70. ftnint cirec;
  71. } cilist;
  72. /*internal read, write*/
  73. typedef struct
  74. { flag icierr;
  75. char *iciunit;
  76. flag iciend;
  77. char *icifmt;
  78. ftnint icirlen;
  79. ftnint icirnum;
  80. } icilist;
  81. /*open*/
  82. typedef struct
  83. { flag oerr;
  84. ftnint ounit;
  85. char *ofnm;
  86. ftnlen ofnmlen;
  87. char *osta;
  88. char *oacc;
  89. char *ofm;
  90. ftnint orl;
  91. char *oblnk;
  92. } olist;
  93. /*close*/
  94. typedef struct
  95. { flag cerr;
  96. ftnint cunit;
  97. char *csta;
  98. } cllist;
  99. /*rewind, backspace, endfile*/
  100. typedef struct
  101. { flag aerr;
  102. ftnint aunit;
  103. } alist;
  104. /* inquire */
  105. typedef struct
  106. { flag inerr;
  107. ftnint inunit;
  108. char *infile;
  109. ftnlen infilen;
  110. ftnint *inex; /*parameters in standard's order*/
  111. ftnint *inopen;
  112. ftnint *innum;
  113. ftnint *innamed;
  114. char *inname;
  115. ftnlen innamlen;
  116. char *inacc;
  117. ftnlen inacclen;
  118. char *inseq;
  119. ftnlen inseqlen;
  120. char *indir;
  121. ftnlen indirlen;
  122. char *infmt;
  123. ftnlen infmtlen;
  124. char *inform;
  125. ftnint informlen;
  126. char *inunf;
  127. ftnlen inunflen;
  128. ftnint *inrecl;
  129. ftnint *innrec;
  130. char *inblank;
  131. ftnlen inblanklen;
  132. } inlist;
  133. #define VOID void
  134. union Multitype { /* for multiple entry points */
  135. integer1 g;
  136. shortint h;
  137. integer i;
  138. /* longint j; */
  139. real r;
  140. doublereal d;
  141. complex c;
  142. doublecomplex z;
  143. };
  144. typedef union Multitype Multitype;
  145. struct Vardesc { /* for Namelist */
  146. char *name;
  147. char *addr;
  148. ftnlen *dims;
  149. int type;
  150. };
  151. typedef struct Vardesc Vardesc;
  152. struct Namelist {
  153. char *name;
  154. Vardesc **vars;
  155. int nvars;
  156. };
  157. typedef struct Namelist Namelist;
  158. #define abs(x) ((x) >= 0 ? (x) : -(x))
  159. #define dabs(x) (fabs(x))
  160. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  161. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  162. #define dmin(a,b) (f2cmin(a,b))
  163. #define dmax(a,b) (f2cmax(a,b))
  164. #define bit_test(a,b) ((a) >> (b) & 1)
  165. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  166. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  167. #define abort_() { sig_die("Fortran abort routine called", 1); }
  168. #define c_abs(z) (cabsf(Cf(z)))
  169. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  170. #ifdef _MSC_VER
  171. #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
  172. #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
  173. #else
  174. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  175. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  176. #endif
  177. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  178. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  179. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  180. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  181. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  182. #define d_abs(x) (fabs(*(x)))
  183. #define d_acos(x) (acos(*(x)))
  184. #define d_asin(x) (asin(*(x)))
  185. #define d_atan(x) (atan(*(x)))
  186. #define d_atn2(x, y) (atan2(*(x),*(y)))
  187. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  188. #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
  189. #define d_cos(x) (cos(*(x)))
  190. #define d_cosh(x) (cosh(*(x)))
  191. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  192. #define d_exp(x) (exp(*(x)))
  193. #define d_imag(z) (cimag(Cd(z)))
  194. #define r_imag(z) (cimagf(Cf(z)))
  195. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  196. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  197. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  198. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  199. #define d_log(x) (log(*(x)))
  200. #define d_mod(x, y) (fmod(*(x), *(y)))
  201. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  202. #define d_nint(x) u_nint(*(x))
  203. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  204. #define d_sign(a,b) u_sign(*(a),*(b))
  205. #define r_sign(a,b) u_sign(*(a),*(b))
  206. #define d_sin(x) (sin(*(x)))
  207. #define d_sinh(x) (sinh(*(x)))
  208. #define d_sqrt(x) (sqrt(*(x)))
  209. #define d_tan(x) (tan(*(x)))
  210. #define d_tanh(x) (tanh(*(x)))
  211. #define i_abs(x) abs(*(x))
  212. #define i_dnnt(x) ((integer)u_nint(*(x)))
  213. #define i_len(s, n) (n)
  214. #define i_nint(x) ((integer)u_nint(*(x)))
  215. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  216. #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
  217. #define pow_si(B,E) spow_ui(*(B),*(E))
  218. #define pow_ri(B,E) spow_ui(*(B),*(E))
  219. #define pow_di(B,E) dpow_ui(*(B),*(E))
  220. #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
  221. #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
  222. #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
  223. #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
  224. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  225. #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
  226. #define sig_die(s, kill) { exit(1); }
  227. #define s_stop(s, n) {exit(0);}
  228. static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
  229. #define z_abs(z) (cabs(Cd(z)))
  230. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  231. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  232. #define myexit_() break;
  233. #define mycycle() continue;
  234. #define myceiling(w) {ceil(w)}
  235. #define myhuge(w) {HUGE_VAL}
  236. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  237. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  238. /* procedure parameter types for -A and -C++ */
  239. #ifdef __cplusplus
  240. typedef logical (*L_fp)(...);
  241. #else
  242. typedef logical (*L_fp)();
  243. #endif
  244. static float spow_ui(float x, integer n) {
  245. float pow=1.0; unsigned long int u;
  246. if(n != 0) {
  247. if(n < 0) n = -n, x = 1/x;
  248. for(u = n; ; ) {
  249. if(u & 01) pow *= x;
  250. if(u >>= 1) x *= x;
  251. else break;
  252. }
  253. }
  254. return pow;
  255. }
  256. static double dpow_ui(double x, integer n) {
  257. double pow=1.0; unsigned long int u;
  258. if(n != 0) {
  259. if(n < 0) n = -n, x = 1/x;
  260. for(u = n; ; ) {
  261. if(u & 01) pow *= x;
  262. if(u >>= 1) x *= x;
  263. else break;
  264. }
  265. }
  266. return pow;
  267. }
  268. #ifdef _MSC_VER
  269. static _Fcomplex cpow_ui(complex x, integer n) {
  270. complex pow={1.0,0.0}; unsigned long int u;
  271. if(n != 0) {
  272. if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
  273. for(u = n; ; ) {
  274. if(u & 01) pow.r *= x.r, pow.i *= x.i;
  275. if(u >>= 1) x.r *= x.r, x.i *= x.i;
  276. else break;
  277. }
  278. }
  279. _Fcomplex p={pow.r, pow.i};
  280. return p;
  281. }
  282. #else
  283. static _Complex float cpow_ui(_Complex float x, integer n) {
  284. _Complex float pow=1.0; unsigned long int u;
  285. if(n != 0) {
  286. if(n < 0) n = -n, x = 1/x;
  287. for(u = n; ; ) {
  288. if(u & 01) pow *= x;
  289. if(u >>= 1) x *= x;
  290. else break;
  291. }
  292. }
  293. return pow;
  294. }
  295. #endif
  296. #ifdef _MSC_VER
  297. static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
  298. _Dcomplex pow={1.0,0.0}; unsigned long int u;
  299. if(n != 0) {
  300. if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
  301. for(u = n; ; ) {
  302. if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
  303. if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
  304. else break;
  305. }
  306. }
  307. _Dcomplex p = {pow._Val[0], pow._Val[1]};
  308. return p;
  309. }
  310. #else
  311. static _Complex double zpow_ui(_Complex double x, integer n) {
  312. _Complex double pow=1.0; unsigned long int u;
  313. if(n != 0) {
  314. if(n < 0) n = -n, x = 1/x;
  315. for(u = n; ; ) {
  316. if(u & 01) pow *= x;
  317. if(u >>= 1) x *= x;
  318. else break;
  319. }
  320. }
  321. return pow;
  322. }
  323. #endif
  324. static integer pow_ii(integer x, integer n) {
  325. integer pow; unsigned long int u;
  326. if (n <= 0) {
  327. if (n == 0 || x == 1) pow = 1;
  328. else if (x != -1) pow = x == 0 ? 1/x : 0;
  329. else n = -n;
  330. }
  331. if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
  332. u = n;
  333. for(pow = 1; ; ) {
  334. if(u & 01) pow *= x;
  335. if(u >>= 1) x *= x;
  336. else break;
  337. }
  338. }
  339. return pow;
  340. }
  341. static integer dmaxloc_(double *w, integer s, integer e, integer *n)
  342. {
  343. double m; integer i, mi;
  344. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  345. if (w[i-1]>m) mi=i ,m=w[i-1];
  346. return mi-s+1;
  347. }
  348. static integer smaxloc_(float *w, integer s, integer e, integer *n)
  349. {
  350. float m; integer i, mi;
  351. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  352. if (w[i-1]>m) mi=i ,m=w[i-1];
  353. return mi-s+1;
  354. }
  355. static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  356. integer n = *n_, incx = *incx_, incy = *incy_, i;
  357. #ifdef _MSC_VER
  358. _Fcomplex zdotc = {0.0, 0.0};
  359. if (incx == 1 && incy == 1) {
  360. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  361. zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
  362. zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
  363. }
  364. } else {
  365. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  366. zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
  367. zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
  368. }
  369. }
  370. pCf(z) = zdotc;
  371. }
  372. #else
  373. _Complex float zdotc = 0.0;
  374. if (incx == 1 && incy == 1) {
  375. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  376. zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
  377. }
  378. } else {
  379. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  380. zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
  381. }
  382. }
  383. pCf(z) = zdotc;
  384. }
  385. #endif
  386. static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  387. integer n = *n_, incx = *incx_, incy = *incy_, i;
  388. #ifdef _MSC_VER
  389. _Dcomplex zdotc = {0.0, 0.0};
  390. if (incx == 1 && incy == 1) {
  391. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  392. zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
  393. zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
  394. }
  395. } else {
  396. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  397. zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
  398. zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
  399. }
  400. }
  401. pCd(z) = zdotc;
  402. }
  403. #else
  404. _Complex double zdotc = 0.0;
  405. if (incx == 1 && incy == 1) {
  406. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  407. zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
  408. }
  409. } else {
  410. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  411. zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
  412. }
  413. }
  414. pCd(z) = zdotc;
  415. }
  416. #endif
  417. static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  418. integer n = *n_, incx = *incx_, incy = *incy_, i;
  419. #ifdef _MSC_VER
  420. _Fcomplex zdotc = {0.0, 0.0};
  421. if (incx == 1 && incy == 1) {
  422. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  423. zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
  424. zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
  425. }
  426. } else {
  427. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  428. zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
  429. zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
  430. }
  431. }
  432. pCf(z) = zdotc;
  433. }
  434. #else
  435. _Complex float zdotc = 0.0;
  436. if (incx == 1 && incy == 1) {
  437. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  438. zdotc += Cf(&x[i]) * Cf(&y[i]);
  439. }
  440. } else {
  441. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  442. zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
  443. }
  444. }
  445. pCf(z) = zdotc;
  446. }
  447. #endif
  448. static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  449. integer n = *n_, incx = *incx_, incy = *incy_, i;
  450. #ifdef _MSC_VER
  451. _Dcomplex zdotc = {0.0, 0.0};
  452. if (incx == 1 && incy == 1) {
  453. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  454. zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
  455. zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
  456. }
  457. } else {
  458. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  459. zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
  460. zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
  461. }
  462. }
  463. pCd(z) = zdotc;
  464. }
  465. #else
  466. _Complex double zdotc = 0.0;
  467. if (incx == 1 && incy == 1) {
  468. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  469. zdotc += Cd(&x[i]) * Cd(&y[i]);
  470. }
  471. } else {
  472. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  473. zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
  474. }
  475. }
  476. pCd(z) = zdotc;
  477. }
  478. #endif
  479. /* -- translated by f2c (version 20000121).
  480. You must link the resulting object file with the libraries:
  481. -lf2c -lm (in that order)
  482. */
  483. /* Table of constant values */
  484. static integer c__1 = 1;
  485. static real c_b9 = -1.f;
  486. static real c_b11 = 1.f;
  487. /* > \brief \b SLA_PORFSX_EXTENDED improves the computed solution to a system of linear equations for symmetri
  488. c or Hermitian positive-definite matrices by performing extra-precise iterative refinement and provide
  489. s error bounds and backward error estimates fo */
  490. /* r the solution. */
  491. /* =========== DOCUMENTATION =========== */
  492. /* Online html documentation available at */
  493. /* http://www.netlib.org/lapack/explore-html/ */
  494. /* > \htmlonly */
  495. /* > Download SLA_PORFSX_EXTENDED + dependencies */
  496. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/sla_por
  497. fsx_extended.f"> */
  498. /* > [TGZ]</a> */
  499. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/sla_por
  500. fsx_extended.f"> */
  501. /* > [ZIP]</a> */
  502. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/sla_por
  503. fsx_extended.f"> */
  504. /* > [TXT]</a> */
  505. /* > \endhtmlonly */
  506. /* Definition: */
  507. /* =========== */
  508. /* SUBROUTINE SLA_PORFSX_EXTENDED( PREC_TYPE, UPLO, N, NRHS, A, LDA, */
  509. /* AF, LDAF, COLEQU, C, B, LDB, Y, */
  510. /* LDY, BERR_OUT, N_NORMS, */
  511. /* ERR_BNDS_NORM, ERR_BNDS_COMP, RES, */
  512. /* AYB, DY, Y_TAIL, RCOND, ITHRESH, */
  513. /* RTHRESH, DZ_UB, IGNORE_CWISE, */
  514. /* INFO ) */
  515. /* INTEGER INFO, LDA, LDAF, LDB, LDY, N, NRHS, PREC_TYPE, */
  516. /* $ N_NORMS, ITHRESH */
  517. /* CHARACTER UPLO */
  518. /* LOGICAL COLEQU, IGNORE_CWISE */
  519. /* REAL RTHRESH, DZ_UB */
  520. /* REAL A( LDA, * ), AF( LDAF, * ), B( LDB, * ), */
  521. /* $ Y( LDY, * ), RES( * ), DY( * ), Y_TAIL( * ) */
  522. /* REAL C( * ), AYB(*), RCOND, BERR_OUT( * ), */
  523. /* $ ERR_BNDS_NORM( NRHS, * ), */
  524. /* $ ERR_BNDS_COMP( NRHS, * ) */
  525. /* > \par Purpose: */
  526. /* ============= */
  527. /* > */
  528. /* > \verbatim */
  529. /* > */
  530. /* > SLA_PORFSX_EXTENDED improves the computed solution to a system of */
  531. /* > linear equations by performing extra-precise iterative refinement */
  532. /* > and provides error bounds and backward error estimates for the solution. */
  533. /* > This subroutine is called by SPORFSX to perform iterative refinement. */
  534. /* > In addition to normwise error bound, the code provides maximum */
  535. /* > componentwise error bound if possible. See comments for ERR_BNDS_NORM */
  536. /* > and ERR_BNDS_COMP for details of the error bounds. Note that this */
  537. /* > subroutine is only resonsible for setting the second fields of */
  538. /* > ERR_BNDS_NORM and ERR_BNDS_COMP. */
  539. /* > \endverbatim */
  540. /* Arguments: */
  541. /* ========== */
  542. /* > \param[in] PREC_TYPE */
  543. /* > \verbatim */
  544. /* > PREC_TYPE is INTEGER */
  545. /* > Specifies the intermediate precision to be used in refinement. */
  546. /* > The value is defined by ILAPREC(P) where P is a CHARACTER and P */
  547. /* > = 'S': Single */
  548. /* > = 'D': Double */
  549. /* > = 'I': Indigenous */
  550. /* > = 'X' or 'E': Extra */
  551. /* > \endverbatim */
  552. /* > */
  553. /* > \param[in] UPLO */
  554. /* > \verbatim */
  555. /* > UPLO is CHARACTER*1 */
  556. /* > = 'U': Upper triangle of A is stored; */
  557. /* > = 'L': Lower triangle of A is stored. */
  558. /* > \endverbatim */
  559. /* > */
  560. /* > \param[in] N */
  561. /* > \verbatim */
  562. /* > N is INTEGER */
  563. /* > The number of linear equations, i.e., the order of the */
  564. /* > matrix A. N >= 0. */
  565. /* > \endverbatim */
  566. /* > */
  567. /* > \param[in] NRHS */
  568. /* > \verbatim */
  569. /* > NRHS is INTEGER */
  570. /* > The number of right-hand-sides, i.e., the number of columns of the */
  571. /* > matrix B. */
  572. /* > \endverbatim */
  573. /* > */
  574. /* > \param[in] A */
  575. /* > \verbatim */
  576. /* > A is REAL array, dimension (LDA,N) */
  577. /* > On entry, the N-by-N matrix A. */
  578. /* > \endverbatim */
  579. /* > */
  580. /* > \param[in] LDA */
  581. /* > \verbatim */
  582. /* > LDA is INTEGER */
  583. /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
  584. /* > \endverbatim */
  585. /* > */
  586. /* > \param[in] AF */
  587. /* > \verbatim */
  588. /* > AF is REAL array, dimension (LDAF,N) */
  589. /* > The triangular factor U or L from the Cholesky factorization */
  590. /* > A = U**T*U or A = L*L**T, as computed by SPOTRF. */
  591. /* > \endverbatim */
  592. /* > */
  593. /* > \param[in] LDAF */
  594. /* > \verbatim */
  595. /* > LDAF is INTEGER */
  596. /* > The leading dimension of the array AF. LDAF >= f2cmax(1,N). */
  597. /* > \endverbatim */
  598. /* > */
  599. /* > \param[in] COLEQU */
  600. /* > \verbatim */
  601. /* > COLEQU is LOGICAL */
  602. /* > If .TRUE. then column equilibration was done to A before calling */
  603. /* > this routine. This is needed to compute the solution and error */
  604. /* > bounds correctly. */
  605. /* > \endverbatim */
  606. /* > */
  607. /* > \param[in] C */
  608. /* > \verbatim */
  609. /* > C is REAL array, dimension (N) */
  610. /* > The column scale factors for A. If COLEQU = .FALSE., C */
  611. /* > is not accessed. If C is input, each element of C should be a power */
  612. /* > of the radix to ensure a reliable solution and error estimates. */
  613. /* > Scaling by powers of the radix does not cause rounding errors unless */
  614. /* > the result underflows or overflows. Rounding errors during scaling */
  615. /* > lead to refining with a matrix that is not equivalent to the */
  616. /* > input matrix, producing error estimates that may not be */
  617. /* > reliable. */
  618. /* > \endverbatim */
  619. /* > */
  620. /* > \param[in] B */
  621. /* > \verbatim */
  622. /* > B is REAL array, dimension (LDB,NRHS) */
  623. /* > The right-hand-side matrix B. */
  624. /* > \endverbatim */
  625. /* > */
  626. /* > \param[in] LDB */
  627. /* > \verbatim */
  628. /* > LDB is INTEGER */
  629. /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
  630. /* > \endverbatim */
  631. /* > */
  632. /* > \param[in,out] Y */
  633. /* > \verbatim */
  634. /* > Y is REAL array, dimension (LDY,NRHS) */
  635. /* > On entry, the solution matrix X, as computed by SPOTRS. */
  636. /* > On exit, the improved solution matrix Y. */
  637. /* > \endverbatim */
  638. /* > */
  639. /* > \param[in] LDY */
  640. /* > \verbatim */
  641. /* > LDY is INTEGER */
  642. /* > The leading dimension of the array Y. LDY >= f2cmax(1,N). */
  643. /* > \endverbatim */
  644. /* > */
  645. /* > \param[out] BERR_OUT */
  646. /* > \verbatim */
  647. /* > BERR_OUT is REAL array, dimension (NRHS) */
  648. /* > On exit, BERR_OUT(j) contains the componentwise relative backward */
  649. /* > error for right-hand-side j from the formula */
  650. /* > f2cmax(i) ( abs(RES(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) ) */
  651. /* > where abs(Z) is the componentwise absolute value of the matrix */
  652. /* > or vector Z. This is computed by SLA_LIN_BERR. */
  653. /* > \endverbatim */
  654. /* > */
  655. /* > \param[in] N_NORMS */
  656. /* > \verbatim */
  657. /* > N_NORMS is INTEGER */
  658. /* > Determines which error bounds to return (see ERR_BNDS_NORM */
  659. /* > and ERR_BNDS_COMP). */
  660. /* > If N_NORMS >= 1 return normwise error bounds. */
  661. /* > If N_NORMS >= 2 return componentwise error bounds. */
  662. /* > \endverbatim */
  663. /* > */
  664. /* > \param[in,out] ERR_BNDS_NORM */
  665. /* > \verbatim */
  666. /* > ERR_BNDS_NORM is REAL array, dimension (NRHS, N_ERR_BNDS) */
  667. /* > For each right-hand side, this array contains information about */
  668. /* > various error bounds and condition numbers corresponding to the */
  669. /* > normwise relative error, which is defined as follows: */
  670. /* > */
  671. /* > Normwise relative error in the ith solution vector: */
  672. /* > max_j (abs(XTRUE(j,i) - X(j,i))) */
  673. /* > ------------------------------ */
  674. /* > max_j abs(X(j,i)) */
  675. /* > */
  676. /* > The array is indexed by the type of error information as described */
  677. /* > below. There currently are up to three pieces of information */
  678. /* > returned. */
  679. /* > */
  680. /* > The first index in ERR_BNDS_NORM(i,:) corresponds to the ith */
  681. /* > right-hand side. */
  682. /* > */
  683. /* > The second index in ERR_BNDS_NORM(:,err) contains the following */
  684. /* > three fields: */
  685. /* > err = 1 "Trust/don't trust" boolean. Trust the answer if the */
  686. /* > reciprocal condition number is less than the threshold */
  687. /* > sqrt(n) * slamch('Epsilon'). */
  688. /* > */
  689. /* > err = 2 "Guaranteed" error bound: The estimated forward error, */
  690. /* > almost certainly within a factor of 10 of the true error */
  691. /* > so long as the next entry is greater than the threshold */
  692. /* > sqrt(n) * slamch('Epsilon'). This error bound should only */
  693. /* > be trusted if the previous boolean is true. */
  694. /* > */
  695. /* > err = 3 Reciprocal condition number: Estimated normwise */
  696. /* > reciprocal condition number. Compared with the threshold */
  697. /* > sqrt(n) * slamch('Epsilon') to determine if the error */
  698. /* > estimate is "guaranteed". These reciprocal condition */
  699. /* > numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some */
  700. /* > appropriately scaled matrix Z. */
  701. /* > Let Z = S*A, where S scales each row by a power of the */
  702. /* > radix so all absolute row sums of Z are approximately 1. */
  703. /* > */
  704. /* > This subroutine is only responsible for setting the second field */
  705. /* > above. */
  706. /* > See Lapack Working Note 165 for further details and extra */
  707. /* > cautions. */
  708. /* > \endverbatim */
  709. /* > */
  710. /* > \param[in,out] ERR_BNDS_COMP */
  711. /* > \verbatim */
  712. /* > ERR_BNDS_COMP is REAL array, dimension (NRHS, N_ERR_BNDS) */
  713. /* > For each right-hand side, this array contains information about */
  714. /* > various error bounds and condition numbers corresponding to the */
  715. /* > componentwise relative error, which is defined as follows: */
  716. /* > */
  717. /* > Componentwise relative error in the ith solution vector: */
  718. /* > abs(XTRUE(j,i) - X(j,i)) */
  719. /* > max_j ---------------------- */
  720. /* > abs(X(j,i)) */
  721. /* > */
  722. /* > The array is indexed by the right-hand side i (on which the */
  723. /* > componentwise relative error depends), and the type of error */
  724. /* > information as described below. There currently are up to three */
  725. /* > pieces of information returned for each right-hand side. If */
  726. /* > componentwise accuracy is not requested (PARAMS(3) = 0.0), then */
  727. /* > ERR_BNDS_COMP is not accessed. If N_ERR_BNDS < 3, then at most */
  728. /* > the first (:,N_ERR_BNDS) entries are returned. */
  729. /* > */
  730. /* > The first index in ERR_BNDS_COMP(i,:) corresponds to the ith */
  731. /* > right-hand side. */
  732. /* > */
  733. /* > The second index in ERR_BNDS_COMP(:,err) contains the following */
  734. /* > three fields: */
  735. /* > err = 1 "Trust/don't trust" boolean. Trust the answer if the */
  736. /* > reciprocal condition number is less than the threshold */
  737. /* > sqrt(n) * slamch('Epsilon'). */
  738. /* > */
  739. /* > err = 2 "Guaranteed" error bound: The estimated forward error, */
  740. /* > almost certainly within a factor of 10 of the true error */
  741. /* > so long as the next entry is greater than the threshold */
  742. /* > sqrt(n) * slamch('Epsilon'). This error bound should only */
  743. /* > be trusted if the previous boolean is true. */
  744. /* > */
  745. /* > err = 3 Reciprocal condition number: Estimated componentwise */
  746. /* > reciprocal condition number. Compared with the threshold */
  747. /* > sqrt(n) * slamch('Epsilon') to determine if the error */
  748. /* > estimate is "guaranteed". These reciprocal condition */
  749. /* > numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some */
  750. /* > appropriately scaled matrix Z. */
  751. /* > Let Z = S*(A*diag(x)), where x is the solution for the */
  752. /* > current right-hand side and S scales each row of */
  753. /* > A*diag(x) by a power of the radix so all absolute row */
  754. /* > sums of Z are approximately 1. */
  755. /* > */
  756. /* > This subroutine is only responsible for setting the second field */
  757. /* > above. */
  758. /* > See Lapack Working Note 165 for further details and extra */
  759. /* > cautions. */
  760. /* > \endverbatim */
  761. /* > */
  762. /* > \param[in] RES */
  763. /* > \verbatim */
  764. /* > RES is REAL array, dimension (N) */
  765. /* > Workspace to hold the intermediate residual. */
  766. /* > \endverbatim */
  767. /* > */
  768. /* > \param[in] AYB */
  769. /* > \verbatim */
  770. /* > AYB is REAL array, dimension (N) */
  771. /* > Workspace. This can be the same workspace passed for Y_TAIL. */
  772. /* > \endverbatim */
  773. /* > */
  774. /* > \param[in] DY */
  775. /* > \verbatim */
  776. /* > DY is REAL array, dimension (N) */
  777. /* > Workspace to hold the intermediate solution. */
  778. /* > \endverbatim */
  779. /* > */
  780. /* > \param[in] Y_TAIL */
  781. /* > \verbatim */
  782. /* > Y_TAIL is REAL array, dimension (N) */
  783. /* > Workspace to hold the trailing bits of the intermediate solution. */
  784. /* > \endverbatim */
  785. /* > */
  786. /* > \param[in] RCOND */
  787. /* > \verbatim */
  788. /* > RCOND is REAL */
  789. /* > Reciprocal scaled condition number. This is an estimate of the */
  790. /* > reciprocal Skeel condition number of the matrix A after */
  791. /* > equilibration (if done). If this is less than the machine */
  792. /* > precision (in particular, if it is zero), the matrix is singular */
  793. /* > to working precision. Note that the error may still be small even */
  794. /* > if this number is very small and the matrix appears ill- */
  795. /* > conditioned. */
  796. /* > \endverbatim */
  797. /* > */
  798. /* > \param[in] ITHRESH */
  799. /* > \verbatim */
  800. /* > ITHRESH is INTEGER */
  801. /* > The maximum number of residual computations allowed for */
  802. /* > refinement. The default is 10. For 'aggressive' set to 100 to */
  803. /* > permit convergence using approximate factorizations or */
  804. /* > factorizations other than LU. If the factorization uses a */
  805. /* > technique other than Gaussian elimination, the guarantees in */
  806. /* > ERR_BNDS_NORM and ERR_BNDS_COMP may no longer be trustworthy. */
  807. /* > \endverbatim */
  808. /* > */
  809. /* > \param[in] RTHRESH */
  810. /* > \verbatim */
  811. /* > RTHRESH is REAL */
  812. /* > Determines when to stop refinement if the error estimate stops */
  813. /* > decreasing. Refinement will stop when the next solution no longer */
  814. /* > satisfies norm(dx_{i+1}) < RTHRESH * norm(dx_i) where norm(Z) is */
  815. /* > the infinity norm of Z. RTHRESH satisfies 0 < RTHRESH <= 1. The */
  816. /* > default value is 0.5. For 'aggressive' set to 0.9 to permit */
  817. /* > convergence on extremely ill-conditioned matrices. See LAWN 165 */
  818. /* > for more details. */
  819. /* > \endverbatim */
  820. /* > */
  821. /* > \param[in] DZ_UB */
  822. /* > \verbatim */
  823. /* > DZ_UB is REAL */
  824. /* > Determines when to start considering componentwise convergence. */
  825. /* > Componentwise convergence is only considered after each component */
  826. /* > of the solution Y is stable, which we definte as the relative */
  827. /* > change in each component being less than DZ_UB. The default value */
  828. /* > is 0.25, requiring the first bit to be stable. See LAWN 165 for */
  829. /* > more details. */
  830. /* > \endverbatim */
  831. /* > */
  832. /* > \param[in] IGNORE_CWISE */
  833. /* > \verbatim */
  834. /* > IGNORE_CWISE is LOGICAL */
  835. /* > If .TRUE. then ignore componentwise convergence. Default value */
  836. /* > is .FALSE.. */
  837. /* > \endverbatim */
  838. /* > */
  839. /* > \param[out] INFO */
  840. /* > \verbatim */
  841. /* > INFO is INTEGER */
  842. /* > = 0: Successful exit. */
  843. /* > < 0: if INFO = -i, the ith argument to SPOTRS had an illegal */
  844. /* > value */
  845. /* > \endverbatim */
  846. /* Authors: */
  847. /* ======== */
  848. /* > \author Univ. of Tennessee */
  849. /* > \author Univ. of California Berkeley */
  850. /* > \author Univ. of Colorado Denver */
  851. /* > \author NAG Ltd. */
  852. /* > \date December 2016 */
  853. /* > \ingroup realPOcomputational */
  854. /* ===================================================================== */
  855. /* Subroutine */ void sla_porfsx_extended_(integer *prec_type__, char *uplo,
  856. integer *n, integer *nrhs, real *a, integer *lda, real *af, integer *
  857. ldaf, logical *colequ, real *c__, real *b, integer *ldb, real *y,
  858. integer *ldy, real *berr_out__, integer *n_norms__, real *
  859. err_bnds_norm__, real *err_bnds_comp__, real *res, real *ayb, real *
  860. dy, real *y_tail__, real *rcond, integer *ithresh, real *rthresh,
  861. real *dz_ub__, logical *ignore_cwise__, integer *info)
  862. {
  863. /* System generated locals */
  864. integer a_dim1, a_offset, af_dim1, af_offset, b_dim1, b_offset, y_dim1,
  865. y_offset, err_bnds_norm_dim1, err_bnds_norm_offset,
  866. err_bnds_comp_dim1, err_bnds_comp_offset, i__1, i__2, i__3;
  867. real r__1, r__2;
  868. /* Local variables */
  869. real dx_x__, dz_z__, ymin;
  870. extern /* Subroutine */ void sla_lin_berr_(integer *, integer *, integer *
  871. , real *, real *, real *);
  872. real dxratmax, dzratmax;
  873. integer y_prec_state__, uplo2;
  874. extern /* Subroutine */ void blas_ssymv_x_(integer *, integer *, real *,
  875. real *, integer *, real *, integer *, real *, real *, integer *,
  876. integer *);
  877. integer i__, j;
  878. extern logical lsame_(char *, char *);
  879. real dxrat;
  880. logical incr_prec__;
  881. real dzrat;
  882. extern /* Subroutine */ void blas_ssymv2_x_(integer *, integer *, real *,
  883. real *, integer *, real *, real *, integer *, real *, real *,
  884. integer *, integer *), scopy_(integer *, real *, integer *, real *
  885. , integer *);
  886. real normx, normy;
  887. extern /* Subroutine */ void saxpy_(integer *, real *, real *, integer *,
  888. real *, integer *), sla_syamv_(integer *, integer *, real *,
  889. real *, integer *, real *, integer *, real *, real *, integer *);
  890. real myhugeval, prev_dz_z__;
  891. extern /* Subroutine */ void ssymv_(char *, integer *, real *, real *,
  892. integer *, real *, integer *, real *, real *, integer *);
  893. real yk;
  894. extern real slamch_(char *);
  895. real final_dx_x__, final_dz_z__, normdx;
  896. extern /* Subroutine */ void sla_wwaddw_(integer *, real *, real *, real *
  897. ), spotrs_(char *, integer *, integer *, real *, integer *, real *
  898. , integer *, integer *);
  899. real prevnormdx;
  900. integer cnt;
  901. real dyk, eps;
  902. extern integer ilauplo_(char *);
  903. integer x_state__, z_state__;
  904. real incr_thresh__;
  905. /* -- LAPACK computational routine (version 3.7.0) -- */
  906. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  907. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  908. /* December 2016 */
  909. /* ===================================================================== */
  910. /* Parameter adjustments */
  911. err_bnds_comp_dim1 = *nrhs;
  912. err_bnds_comp_offset = 1 + err_bnds_comp_dim1 * 1;
  913. err_bnds_comp__ -= err_bnds_comp_offset;
  914. err_bnds_norm_dim1 = *nrhs;
  915. err_bnds_norm_offset = 1 + err_bnds_norm_dim1 * 1;
  916. err_bnds_norm__ -= err_bnds_norm_offset;
  917. a_dim1 = *lda;
  918. a_offset = 1 + a_dim1 * 1;
  919. a -= a_offset;
  920. af_dim1 = *ldaf;
  921. af_offset = 1 + af_dim1 * 1;
  922. af -= af_offset;
  923. --c__;
  924. b_dim1 = *ldb;
  925. b_offset = 1 + b_dim1 * 1;
  926. b -= b_offset;
  927. y_dim1 = *ldy;
  928. y_offset = 1 + y_dim1 * 1;
  929. y -= y_offset;
  930. --berr_out__;
  931. --res;
  932. --ayb;
  933. --dy;
  934. --y_tail__;
  935. /* Function Body */
  936. if (*info != 0) {
  937. return;
  938. }
  939. eps = slamch_("Epsilon");
  940. myhugeval = slamch_("Overflow");
  941. /* Force MYHUGEVAL to Inf */
  942. myhugeval *= myhugeval;
  943. /* Using MYHUGEVAL may lead to spurious underflows. */
  944. incr_thresh__ = (real) (*n) * eps;
  945. if (lsame_(uplo, "L")) {
  946. uplo2 = ilauplo_("L");
  947. } else {
  948. uplo2 = ilauplo_("U");
  949. }
  950. i__1 = *nrhs;
  951. for (j = 1; j <= i__1; ++j) {
  952. y_prec_state__ = 1;
  953. if (y_prec_state__ == 2) {
  954. i__2 = *n;
  955. for (i__ = 1; i__ <= i__2; ++i__) {
  956. y_tail__[i__] = 0.f;
  957. }
  958. }
  959. dxrat = 0.f;
  960. dxratmax = 0.f;
  961. dzrat = 0.f;
  962. dzratmax = 0.f;
  963. final_dx_x__ = myhugeval;
  964. final_dz_z__ = myhugeval;
  965. prevnormdx = myhugeval;
  966. prev_dz_z__ = myhugeval;
  967. dz_z__ = myhugeval;
  968. dx_x__ = myhugeval;
  969. x_state__ = 1;
  970. z_state__ = 0;
  971. incr_prec__ = FALSE_;
  972. i__2 = *ithresh;
  973. for (cnt = 1; cnt <= i__2; ++cnt) {
  974. /* Compute residual RES = B_s - op(A_s) * Y, */
  975. /* op(A) = A, A**T, or A**H depending on TRANS (and type). */
  976. scopy_(n, &b[j * b_dim1 + 1], &c__1, &res[1], &c__1);
  977. if (y_prec_state__ == 0) {
  978. ssymv_(uplo, n, &c_b9, &a[a_offset], lda, &y[j * y_dim1 + 1],
  979. &c__1, &c_b11, &res[1], &c__1);
  980. } else if (y_prec_state__ == 1) {
  981. blas_ssymv_x__(&uplo2, n, &c_b9, &a[a_offset], lda, &y[j *
  982. y_dim1 + 1], &c__1, &c_b11, &res[1], &c__1,
  983. prec_type__);
  984. } else {
  985. blas_ssymv2_x__(&uplo2, n, &c_b9, &a[a_offset], lda, &y[j *
  986. y_dim1 + 1], &y_tail__[1], &c__1, &c_b11, &res[1], &
  987. c__1, prec_type__);
  988. }
  989. /* XXX: RES is no longer needed. */
  990. scopy_(n, &res[1], &c__1, &dy[1], &c__1);
  991. spotrs_(uplo, n, &c__1, &af[af_offset], ldaf, &dy[1], n, info);
  992. /* Calculate relative changes DX_X, DZ_Z and ratios DXRAT, DZRAT. */
  993. normx = 0.f;
  994. normy = 0.f;
  995. normdx = 0.f;
  996. dz_z__ = 0.f;
  997. ymin = myhugeval;
  998. i__3 = *n;
  999. for (i__ = 1; i__ <= i__3; ++i__) {
  1000. yk = (r__1 = y[i__ + j * y_dim1], abs(r__1));
  1001. dyk = (r__1 = dy[i__], abs(r__1));
  1002. if (yk != 0.f) {
  1003. /* Computing MAX */
  1004. r__1 = dz_z__, r__2 = dyk / yk;
  1005. dz_z__ = f2cmax(r__1,r__2);
  1006. } else if (dyk != 0.f) {
  1007. dz_z__ = myhugeval;
  1008. }
  1009. ymin = f2cmin(ymin,yk);
  1010. normy = f2cmax(normy,yk);
  1011. if (*colequ) {
  1012. /* Computing MAX */
  1013. r__1 = normx, r__2 = yk * c__[i__];
  1014. normx = f2cmax(r__1,r__2);
  1015. /* Computing MAX */
  1016. r__1 = normdx, r__2 = dyk * c__[i__];
  1017. normdx = f2cmax(r__1,r__2);
  1018. } else {
  1019. normx = normy;
  1020. normdx = f2cmax(normdx,dyk);
  1021. }
  1022. }
  1023. if (normx != 0.f) {
  1024. dx_x__ = normdx / normx;
  1025. } else if (normdx == 0.f) {
  1026. dx_x__ = 0.f;
  1027. } else {
  1028. dx_x__ = myhugeval;
  1029. }
  1030. dxrat = normdx / prevnormdx;
  1031. dzrat = dz_z__ / prev_dz_z__;
  1032. /* Check termination criteria. */
  1033. if (ymin * *rcond < incr_thresh__ * normy && y_prec_state__ < 2) {
  1034. incr_prec__ = TRUE_;
  1035. }
  1036. if (x_state__ == 3 && dxrat <= *rthresh) {
  1037. x_state__ = 1;
  1038. }
  1039. if (x_state__ == 1) {
  1040. if (dx_x__ <= eps) {
  1041. x_state__ = 2;
  1042. } else if (dxrat > *rthresh) {
  1043. if (y_prec_state__ != 2) {
  1044. incr_prec__ = TRUE_;
  1045. } else {
  1046. x_state__ = 3;
  1047. }
  1048. } else {
  1049. if (dxrat > dxratmax) {
  1050. dxratmax = dxrat;
  1051. }
  1052. }
  1053. if (x_state__ > 1) {
  1054. final_dx_x__ = dx_x__;
  1055. }
  1056. }
  1057. if (z_state__ == 0 && dz_z__ <= *dz_ub__) {
  1058. z_state__ = 1;
  1059. }
  1060. if (z_state__ == 3 && dzrat <= *rthresh) {
  1061. z_state__ = 1;
  1062. }
  1063. if (z_state__ == 1) {
  1064. if (dz_z__ <= eps) {
  1065. z_state__ = 2;
  1066. } else if (dz_z__ > *dz_ub__) {
  1067. z_state__ = 0;
  1068. dzratmax = 0.f;
  1069. final_dz_z__ = myhugeval;
  1070. } else if (dzrat > *rthresh) {
  1071. if (y_prec_state__ != 2) {
  1072. incr_prec__ = TRUE_;
  1073. } else {
  1074. z_state__ = 3;
  1075. }
  1076. } else {
  1077. if (dzrat > dzratmax) {
  1078. dzratmax = dzrat;
  1079. }
  1080. }
  1081. if (z_state__ > 1) {
  1082. final_dz_z__ = dz_z__;
  1083. }
  1084. }
  1085. if (x_state__ != 1 && (*ignore_cwise__ || z_state__ != 1)) {
  1086. goto L666;
  1087. }
  1088. if (incr_prec__) {
  1089. incr_prec__ = FALSE_;
  1090. ++y_prec_state__;
  1091. i__3 = *n;
  1092. for (i__ = 1; i__ <= i__3; ++i__) {
  1093. y_tail__[i__] = 0.f;
  1094. }
  1095. }
  1096. prevnormdx = normdx;
  1097. prev_dz_z__ = dz_z__;
  1098. /* Update soluton. */
  1099. if (y_prec_state__ < 2) {
  1100. saxpy_(n, &c_b11, &dy[1], &c__1, &y[j * y_dim1 + 1], &c__1);
  1101. } else {
  1102. sla_wwaddw_(n, &y[j * y_dim1 + 1], &y_tail__[1], &dy[1]);
  1103. }
  1104. }
  1105. /* Target of "IF (Z_STOP .AND. X_STOP)". Sun's f77 won't CALL MYEXIT. */
  1106. L666:
  1107. /* Set final_* when cnt hits ithresh. */
  1108. if (x_state__ == 1) {
  1109. final_dx_x__ = dx_x__;
  1110. }
  1111. if (z_state__ == 1) {
  1112. final_dz_z__ = dz_z__;
  1113. }
  1114. /* Compute error bounds. */
  1115. if (*n_norms__ >= 1) {
  1116. err_bnds_norm__[j + (err_bnds_norm_dim1 << 1)] = final_dx_x__ / (
  1117. 1 - dxratmax);
  1118. }
  1119. if (*n_norms__ >= 2) {
  1120. err_bnds_comp__[j + (err_bnds_comp_dim1 << 1)] = final_dz_z__ / (
  1121. 1 - dzratmax);
  1122. }
  1123. /* Compute componentwise relative backward error from formula */
  1124. /* f2cmax(i) ( abs(R(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) ) */
  1125. /* where abs(Z) is the componentwise absolute value of the matrix */
  1126. /* or vector Z. */
  1127. /* Compute residual RES = B_s - op(A_s) * Y, */
  1128. /* op(A) = A, A**T, or A**H depending on TRANS (and type). */
  1129. scopy_(n, &b[j * b_dim1 + 1], &c__1, &res[1], &c__1);
  1130. ssymv_(uplo, n, &c_b9, &a[a_offset], lda, &y[j * y_dim1 + 1], &c__1, &
  1131. c_b11, &res[1], &c__1);
  1132. i__2 = *n;
  1133. for (i__ = 1; i__ <= i__2; ++i__) {
  1134. ayb[i__] = (r__1 = b[i__ + j * b_dim1], abs(r__1));
  1135. }
  1136. /* Compute abs(op(A_s))*abs(Y) + abs(B_s). */
  1137. sla_syamv_(&uplo2, n, &c_b11, &a[a_offset], lda, &y[j * y_dim1 + 1],
  1138. &c__1, &c_b11, &ayb[1], &c__1);
  1139. sla_lin_berr_(n, n, &c__1, &res[1], &ayb[1], &berr_out__[j]);
  1140. /* End of loop for each RHS. */
  1141. }
  1142. return;
  1143. } /* sla_porfsx_extended__ */